基于MHSA和C-BiGRU的航班延误预测研究  

Flight Delay Prediction Research Based on MHSA and C-BiGRU

作　　者：黄程 王辉[1] 吴俊霖 王鑫玮 檀萝帆 HUANG Cheng;WANG Hui;WU Jun-lin;WANG Xin-wei;TAN Luo-fan(Civil Aviation University of China,Tianjin 300000,China;Shanghai Airport Group,Shanghai 200000,China)

机构地区：[1]中国民航大学,天津300000 [2]上海机场集团,上海200000

出　　处：《航空计算技术》2023年第6期65-69,共5页Aeronautical Computing Technique

基　　金：上海机场集团科研合作项目资助(H04420220072)。

摘　　要：针对普通神经网络预测人为因素造成的航班延误能力不足的问题,提出了一种基于多头自注意力机制和卷积双向门控循环单元的预测模型(MHSA-C-BiGRU)对航班延误问题进行研究。模型采用卷积双向门控循环单元(C-BiGRU)提取局部信息和上下游数据的时序信息,利用多头自注意力机制(MHSA)的并行能力从不同位置提取数据内部之间的特征,强化重要信息的权值,使模型聚焦到对当前任务更重要的信息,从而增强模型分析人为因素造成的航班延误的能力。研究使用2018年上海浦东机场的航班数据和气象数据。结果表明,预测模型相对于基础模型的预测准确率提高了4.4%,各项宏平均值有8%的提高,各项权重平均值有5%的提高。Aiming at the insufficient ability of common neural networks to predict flight delays caused by human factors,this paper proposes a prediction model based on the multi-head self-attention mechanism and convolutional bidirectional gated recurrent unit(MHSA-C-BiGRU)to study the flight delay problem.The model uses the convolutional bidirectional gated recurrent unit(C-BiGRU)to extract the local information and temporal information of the upstream and downstream data,and uses the parallel ability of the multi-head self-attention mechanism(MHSA)to extract the features between the data from different locations,and enhances the weights of the important information so that the model focuses on the information that is more important for the current task,thus improving the model′s ability to analyze the flight delays caused by human factors.This study uses flight data and meteorological data from Shanghai Pudong Airport in 2018.The results show that the prediction accuracy of the prediction model proposed in this paper is improved by 4.4%compared to the base model,there is an 8%improvement in the average value of each macro,and there is a 5%improvement in the average value of each weight.

关 键 词：航班延误预测 多头自注意力机制 卷积双向门控循环单元 时空特征 

分 类 号：V355[航空宇航科学与技术—人机与环境工程]